/* Samplers */
unsigned num_samplers;
+ boolean sampler_view[PIPE_MAX_SAMPLERS]; /**< True if sampler view exists*/
+ ubyte sampler_target[PIPE_MAX_SAMPLERS]; /**< TGSI_TEXTURE_x */
+ ubyte sampler_return_type[PIPE_MAX_SAMPLERS]; /**< TGSI_RETURN_TYPE_x */
/* Address regs (really implemented with temps) */
unsigned num_address_regs;
/* For fragment shaders only */
struct {
- /* apha test */
unsigned color_out_index[PIPE_MAX_COLOR_BUFS]; /**< the real color output regs */
+ unsigned num_color_outputs;
unsigned color_tmp_index; /**< fake/temp color output reg */
unsigned alpha_ref_index; /**< immediate constant for alpha ref */
unsigned fragcoord_input_index; /**< real fragment position input reg */
unsigned fragcoord_tmp_index; /**< 1/w modified position temp reg */
+
+ /** Which texture units are doing shadow comparison in the FS code */
+ unsigned shadow_compare_units;
} fs;
/* For geometry shaders only */
/* user clip plane constant slot indexes */
unsigned clip_plane_const[PIPE_MAX_CLIP_PLANES];
+ unsigned num_output_writes;
+ boolean constant_color_output;
+
boolean uses_flat_interp;
/* For all shaders: const reg index for RECT coord scaling */
else
new_buf = NULL;
- if (new_buf == NULL) {
+ if (!new_buf) {
emit->ptr = err_buf;
emit->buf = err_buf;
emit->size = sizeof(err_buf);
* as seen below. All other opcodes are handled/translated specially.
*/
static VGPU10_OPCODE_TYPE
-translate_opcode(unsigned opcode)
+translate_opcode(enum tgsi_opcode opcode)
{
switch (opcode) {
case TGSI_OPCODE_MOV:
return VGPU10_OPCODE_RET;
case TGSI_OPCODE_NOP:
return VGPU10_OPCODE_NOP;
- case TGSI_OPCODE_BREAKC:
- return VGPU10_OPCODE_BREAKC;
case TGSI_OPCODE_END:
return VGPU10_OPCODE_RET;
case TGSI_OPCODE_F2I:
*/
static unsigned
get_temp_array_id(const struct svga_shader_emitter_v10 *emit,
- unsigned file, unsigned index)
+ enum tgsi_file_type file, unsigned index)
{
if (file == TGSI_FILE_TEMPORARY) {
return emit->temp_map[index].arrayId;
*/
static unsigned
remap_temp_index(const struct svga_shader_emitter_v10 *emit,
- unsigned file, unsigned index)
+ enum tgsi_file_type file, unsigned index)
{
if (file == TGSI_FILE_TEMPORARY) {
return emit->temp_map[index].index;
static VGPU10OperandToken0
setup_operand0_indexing(struct svga_shader_emitter_v10 *emit,
VGPU10OperandToken0 operand0,
- unsigned file,
+ enum tgsi_file_type file,
boolean indirect, boolean index2D,
unsigned tempArrayID)
{
- unsigned indexDim, index0Rep, index1Rep = VGPU10_OPERAND_INDEX_0D;
+ unsigned indexDim, index0Rep, index1Rep = VGPU10_OPERAND_INDEX_IMMEDIATE32;
/*
* Compute index dimensions
emit_dst_register(struct svga_shader_emitter_v10 *emit,
const struct tgsi_full_dst_register *reg)
{
- unsigned file = reg->Register.File;
+ enum tgsi_file_type file = reg->Register.File;
unsigned index = reg->Register.Index;
- const unsigned sem_name = emit->info.output_semantic_name[index];
+ const enum tgsi_semantic sem_name = emit->info.output_semantic_name[index];
const unsigned sem_index = emit->info.output_semantic_index[index];
unsigned writemask = reg->Register.WriteMask;
- const unsigned indirect = reg->Register.Indirect;
+ const boolean indirect = reg->Register.Indirect;
const unsigned tempArrayId = get_temp_array_id(emit, file, index);
- const unsigned index2d = reg->Register.Dimension;
+ const boolean index2d = reg->Register.Dimension;
VGPU10OperandToken0 operand0;
if (file == TGSI_FILE_OUTPUT) {
*/
assert(sem_name == TGSI_SEMANTIC_COLOR);
index = emit->info.output_semantic_index[index];
+
+ emit->num_output_writes++;
}
}
}
emit_src_register(struct svga_shader_emitter_v10 *emit,
const struct tgsi_full_src_register *reg)
{
- unsigned file = reg->Register.File;
+ enum tgsi_file_type file = reg->Register.File;
unsigned index = reg->Register.Index;
- const unsigned indirect = reg->Register.Indirect;
+ const boolean indirect = reg->Register.Indirect;
const unsigned tempArrayId = get_temp_array_id(emit, file, index);
- const unsigned index2d = reg->Register.Dimension;
+ const boolean index2d = reg->Register.Dimension;
const unsigned swizzleX = reg->Register.SwizzleX;
const unsigned swizzleY = reg->Register.SwizzleY;
const unsigned swizzleZ = reg->Register.SwizzleZ;
const unsigned swizzleW = reg->Register.SwizzleW;
- const unsigned absolute = reg->Register.Absolute;
- const unsigned negate = reg->Register.Negate;
+ const boolean absolute = reg->Register.Absolute;
+ const boolean negate = reg->Register.Negate;
bool is_prim_id = FALSE;
VGPU10OperandToken0 operand0;
}
}
else if (file == TGSI_FILE_SYSTEM_VALUE) {
- assert(index < Elements(emit->system_value_indexes));
+ assert(index < ARRAY_SIZE(emit->system_value_indexes));
index = emit->system_value_indexes[index];
}
}
operand0.value = operand1.value = 0;
if (is_prim_id) {
+ /* NOTE: we should be using VGPU10_OPERAND_1_COMPONENT here, but
+ * our virtual GPU accepts this as-is.
+ */
operand0.numComponents = VGPU10_OPERAND_0_COMPONENT;
operand0.operandType = VGPU10_OPERAND_TYPE_INPUT_PRIMITIVEID;
}
if (operand0.operandType == VGPU10_OPERAND_TYPE_IMMEDIATE32) {
/* Emit the four float/int in-line immediate values */
unsigned *c;
- assert(index < Elements(emit->immediates));
+ assert(index < ARRAY_SIZE(emit->immediates));
assert(file == TGSI_FILE_IMMEDIATE);
assert(swizzleX < 4);
assert(swizzleY < 4);
*/
static void
emit_opcode(struct svga_shader_emitter_v10 *emit,
- unsigned vgpu10_opcode, boolean saturate)
+ VGPU10_OPCODE_TYPE vgpu10_opcode, boolean saturate)
{
VGPU10OpcodeToken0 token0;
* Create a tgsi_full_src_register.
*/
static struct tgsi_full_src_register
-make_src_reg(unsigned file, unsigned index)
+make_src_reg(enum tgsi_file_type file, unsigned index)
{
struct tgsi_full_src_register reg;
* Create a tgsi_full_dst_register.
*/
static struct tgsi_full_dst_register
-make_dst_reg(unsigned file, unsigned index)
+make_dst_reg(enum tgsi_file_type file, unsigned index)
{
struct tgsi_full_dst_register reg;
/** Return the named swizzle term from the src register */
static inline unsigned
-get_swizzle(const struct tgsi_full_src_register *reg, unsigned term)
+get_swizzle(const struct tgsi_full_src_register *reg, enum tgsi_swizzle term)
{
switch (term) {
case TGSI_SWIZZLE_X:
*/
static struct tgsi_full_src_register
swizzle_src(const struct tgsi_full_src_register *reg,
- unsigned swizzleX, unsigned swizzleY,
- unsigned swizzleZ, unsigned swizzleW)
+ enum tgsi_swizzle swizzleX, enum tgsi_swizzle swizzleY,
+ enum tgsi_swizzle swizzleZ, enum tgsi_swizzle swizzleW)
{
struct tgsi_full_src_register swizzled = *reg;
/* Note: we swizzle the current swizzle */
* terms are the same.
*/
static struct tgsi_full_src_register
-scalar_src(const struct tgsi_full_src_register *reg, unsigned swizzle)
+scalar_src(const struct tgsi_full_src_register *reg, enum tgsi_swizzle swizzle)
{
struct tgsi_full_src_register swizzled = *reg;
/* Note: we swizzle the current swizzle */
{
unsigned n = emit->num_immediates++;
assert(!emit->immediates_emitted);
- assert(n < Elements(emit->immediates));
+ assert(n < ARRAY_SIZE(emit->immediates));
emit->immediates[n][0] = imm[0];
emit->immediates[n][1] = imm[1];
emit->immediates[n][2] = imm[2];
/**
- * Allocate space for a int[4] immediate.
+ * Allocate space for an int[4] immediate.
* \return the index/position of the immediate.
*/
static unsigned
static unsigned
alloc_system_value_index(struct svga_shader_emitter_v10 *emit, unsigned index)
{
- const unsigned n = emit->info.num_inputs + index;
- assert(index < Elements(emit->system_value_indexes));
+ const unsigned n = emit->info.file_max[TGSI_FILE_INPUT] + 1 + index;
+ assert(index < ARRAY_SIZE(emit->system_value_indexes));
emit->system_value_indexes[index] = n;
return n;
}
*/
static unsigned
translate_interpolation(const struct svga_shader_emitter_v10 *emit,
- unsigned interp, unsigned interpolate_loc)
+ enum tgsi_interpolate_mode interp,
+ enum tgsi_interpolate_loc interpolate_loc)
{
if (interp == TGSI_INTERPOLATE_COLOR) {
interp = emit->key.fs.flatshade ?
/**
* Translate a TGSI property to VGPU10.
- * Don't emit any instructions yet, only need to gather the primitive property information.
- * The output primitive topology might be changed later. The final property instructions
- * will be emitted as part of the pre-helper code.
+ * Don't emit any instructions yet, only need to gather the primitive property
+ * information. The output primitive topology might be changed later. The
+ * final property instructions will be emitted as part of the pre-helper code.
*/
static boolean
emit_vgpu10_property(struct svga_shader_emitter_v10 *emit,
switch (prop->Property.PropertyName) {
case TGSI_PROPERTY_GS_INPUT_PRIM:
- assert(prop->u[0].Data < Elements(primType));
+ assert(prop->u[0].Data < ARRAY_SIZE(primType));
emit->gs.prim_type = primType[prop->u[0].Data];
assert(emit->gs.prim_type != VGPU10_PRIMITIVE_UNDEFINED);
emit->gs.input_size = inputArraySize[emit->gs.prim_type];
break;
case TGSI_PROPERTY_GS_OUTPUT_PRIM:
- assert(prop->u[0].Data < Elements(primTopology));
+ assert(prop->u[0].Data < ARRAY_SIZE(primTopology));
emit->gs.prim_topology = primTopology[prop->u[0].Data];
assert(emit->gs.prim_topology != VGPU10_PRIMITIVE_TOPOLOGY_UNDEFINED);
break;
*/
static void
emit_input_declaration(struct svga_shader_emitter_v10 *emit,
- unsigned opcodeType, unsigned operandType,
- unsigned dim, unsigned index, unsigned size,
- unsigned name, unsigned numComp,
- unsigned selMode, unsigned usageMask,
- unsigned interpMode)
+ VGPU10_OPCODE_TYPE opcodeType,
+ VGPU10_OPERAND_TYPE operandType,
+ VGPU10_OPERAND_INDEX_DIMENSION dim,
+ unsigned index, unsigned size,
+ VGPU10_SYSTEM_NAME name,
+ VGPU10_OPERAND_NUM_COMPONENTS numComp,
+ VGPU10_OPERAND_4_COMPONENT_SELECTION_MODE selMode,
+ unsigned usageMask,
+ VGPU10_INTERPOLATION_MODE interpMode)
{
VGPU10OpcodeToken0 opcode0;
VGPU10OperandToken0 operand0;
*/
static void
emit_output_declaration(struct svga_shader_emitter_v10 *emit,
- unsigned type, unsigned index,
- unsigned name, unsigned usageMask)
+ VGPU10_OPCODE_TYPE type, unsigned index,
+ VGPU10_SYSTEM_NAME name,
+ unsigned usageMask)
{
VGPU10OpcodeToken0 opcode0;
VGPU10OperandToken0 operand0;
*/
static void
emit_system_value_declaration(struct svga_shader_emitter_v10 *emit,
- unsigned semantic_name, unsigned index)
+ enum tgsi_semantic semantic_name, unsigned index)
{
switch (semantic_name) {
case TGSI_SEMANTIC_INSTANCEID:
* have linked due to constbuf index out of bounds, so we shouldn't
* have reached here.
*/
- assert(constbuf < Elements(emit->num_shader_consts));
+ assert(constbuf < ARRAY_SIZE(emit->num_shader_consts));
num_consts = MAX2(emit->num_shader_consts[constbuf],
decl->Range.Last + 1);
emit->num_samplers = MAX2(emit->num_samplers, decl->Range.Last + 1);
return TRUE;
+#if 0
case TGSI_FILE_RESOURCE:
/*opcode0.opcodeType = VGPU10_OPCODE_DCL_RESOURCE;*/
/* XXX more, VGPU10_RETURN_TYPE_FLOAT */
assert(!"TGSI_FILE_RESOURCE not handled yet");
return FALSE;
+#endif
case TGSI_FILE_ADDRESS:
emit->num_address_regs = MAX2(emit->num_address_regs,
return TRUE;
case TGSI_FILE_SAMPLER_VIEW:
- /* Not used at this time, but maybe in the future.
- * See emit_resource_declarations().
- */
+ {
+ unsigned unit = decl->Range.First;
+ assert(decl->Range.First == decl->Range.Last);
+ emit->sampler_target[unit] = decl->SamplerView.Resource;
+ /* Note: we can ignore YZW return types for now */
+ emit->sampler_return_type[unit] = decl->SamplerView.ReturnTypeX;
+ emit->sampler_view[unit] = TRUE;
+ }
return TRUE;
default:
if (emit->unit == PIPE_SHADER_FRAGMENT) {
for (i = 0; i < emit->linkage.num_inputs; i++) {
- unsigned semantic_name = emit->info.input_semantic_name[i];
+ enum tgsi_semantic semantic_name = emit->info.input_semantic_name[i];
unsigned usage_mask = emit->info.input_usage_mask[i];
unsigned index = emit->linkage.input_map[i];
- unsigned type, interpolationMode, name;
+ VGPU10_OPCODE_TYPE type;
+ VGPU10_INTERPOLATION_MODE interpolationMode;
+ VGPU10_SYSTEM_NAME name;
if (usage_mask == 0)
continue; /* register is not actually used */
else if (emit->unit == PIPE_SHADER_GEOMETRY) {
for (i = 0; i < emit->info.num_inputs; i++) {
- unsigned semantic_name = emit->info.input_semantic_name[i];
+ enum tgsi_semantic semantic_name = emit->info.input_semantic_name[i];
unsigned usage_mask = emit->info.input_usage_mask[i];
unsigned index = emit->linkage.input_map[i];
- unsigned opcodeType, operandType;
- unsigned numComp, selMode;
- unsigned name;
- unsigned dim;
+ VGPU10_OPCODE_TYPE opcodeType, operandType;
+ VGPU10_OPERAND_NUM_COMPONENTS numComp;
+ VGPU10_OPERAND_4_COMPONENT_SELECTION_MODE selMode;
+ VGPU10_SYSTEM_NAME name;
+ VGPU10_OPERAND_INDEX_DIMENSION dim;
if (usage_mask == 0)
continue; /* register is not actually used */
selMode = VGPU10_OPERAND_4_COMPONENT_MASK_MODE;
name = VGPU10_NAME_UNDEFINED;
- /* all geometry shader inputs are two dimensional except gl_PrimitiveID */
+ /* all geometry shader inputs are two dimensional except
+ * gl_PrimitiveID
+ */
dim = VGPU10_OPERAND_INDEX_2D;
if (semantic_name == TGSI_SEMANTIC_PRIMID) {
else {
assert(emit->unit == PIPE_SHADER_VERTEX);
- for (i = 0; i < emit->info.num_inputs; i++) {
+ for (i = 0; i < emit->info.file_max[TGSI_FILE_INPUT] + 1; i++) {
unsigned usage_mask = emit->info.input_usage_mask[i];
unsigned index = i;
for (i = 0; i < emit->info.num_outputs; i++) {
/*const unsigned usage_mask = emit->info.output_usage_mask[i];*/
- const unsigned semantic_name = emit->info.output_semantic_name[i];
+ const enum tgsi_semantic semantic_name =
+ emit->info.output_semantic_name[i];
const unsigned semantic_index = emit->info.output_semantic_index[i];
unsigned index = i;
if (emit->unit == PIPE_SHADER_FRAGMENT) {
if (semantic_name == TGSI_SEMANTIC_COLOR) {
- assert(semantic_index < Elements(emit->fs.color_out_index));
+ assert(semantic_index < ARRAY_SIZE(emit->fs.color_out_index));
emit->fs.color_out_index[semantic_index] = index;
+ emit->fs.num_color_outputs = MAX2(emit->fs.num_color_outputs,
+ index + 1);
+
/* The semantic index is the shader's color output/buffer index */
emit_output_declaration(emit,
VGPU10_OPCODE_DCL_OUTPUT, semantic_index,
VGPU10_OPERAND_4_COMPONENT_MASK_ALL);
emit->info.output_semantic_index[idx] = j;
}
+
+ emit->fs.num_color_outputs =
+ emit->key.fs.write_color0_to_n_cbufs;
}
}
else {
}
else {
/* VS or GS */
- unsigned name, type;
+ VGPU10_COMPONENT_NAME name;
+ VGPU10_OPCODE_TYPE type;
unsigned writemask = VGPU10_OPERAND_4_COMPONENT_MASK_ALL;
switch (semantic_name) {
total_temps = emit->num_shader_temps;
+ /* If there is indirect access to non-indexable temps in the shader,
+ * convert those temps to indexable temps. This works around a bug
+ * in the GLSL->TGSI translator exposed in piglit test
+ * glsl-1.20/execution/fs-const-array-of-struct-of-array.shader_test.
+ * Internal temps added by the driver remain as non-indexable temps.
+ */
+ if ((emit->info.indirect_files & (1 << TGSI_FILE_TEMPORARY)) &&
+ emit->num_temp_arrays == 0) {
+ unsigned arrayID;
+
+ arrayID = 1;
+ emit->num_temp_arrays = arrayID + 1;
+ emit->temp_arrays[arrayID].start = 0;
+ emit->temp_arrays[arrayID].size = total_temps;
+
+ /* Fill in the temp_map entries for this temp array */
+ for (i = 0; i < total_temps; i++) {
+ emit->temp_map[i].arrayId = arrayID;
+ emit->temp_map[i].index = i;
+ }
+ }
+
/* Allocate extra temps for specially-implemented instructions,
* such as LIT.
*/
emit->temp_map[i].index = reg++;
}
}
- total_temps = reg;
if (0) {
debug_printf("total_temps %u\n", total_temps);
- for (i = 0; i < 30; i++) {
+ for (i = 0; i < total_temps; i++) {
debug_printf("temp %u -> array %u index %u\n",
i, emit->temp_map[i].arrayId, emit->temp_map[i].index);
}
}
+ total_temps = reg;
+
/* Emit declaration of ordinary temp registers */
if (total_temps > 0) {
VGPU10OpcodeToken0 opcode0;
*/
total_consts = emit->num_shader_consts[0];
- /* Now, allocate constant slots for the "extra" constants */
+ /* Now, allocate constant slots for the "extra" constants.
+ * Note: it's critical that these extra constant locations
+ * exactly match what's emitted by the "extra" constants code
+ * in svga_state_constants.c
+ */
/* Vertex position scale/translation */
if (emit->vposition.need_prescale) {
}
}
- /* Texcoord scale factors for RECT textures */
- {
- for (i = 0; i < emit->num_samplers; i++) {
+ for (i = 0; i < emit->num_samplers; i++) {
+
+ if (emit->sampler_view[i]) {
+
+ /* Texcoord scale factors for RECT textures */
if (emit->key.tex[i].unnormalized) {
emit->texcoord_scale_index[i] = total_consts++;
}
- }
- }
- /* Texture buffer sizes */
- for (i = 0; i < emit->num_samplers; i++) {
- if (emit->key.tex[i].texture_target == PIPE_BUFFER) {
- emit->texture_buffer_size_index[i] = total_consts++;
+ /* Texture buffer sizes */
+ if (emit->sampler_target[i] == TGSI_TEXTURE_BUFFER) {
+ emit->texture_buffer_size_index[i] = total_consts++;
+ }
}
}
}
/* Declare remaining constant buffers (UBOs) */
- for (i = 1; i < Elements(emit->num_shader_consts); i++) {
+ for (i = 1; i < ARRAY_SIZE(emit->num_shader_consts); i++) {
if (emit->num_shader_consts[i] > 0) {
begin_emit_instruction(emit);
emit_dword(emit, opcode0.value);
/**
- * Translate PIPE_TEXTURE_x to VGAPU10_RESOURCE_DIMENSION_x.
+ * Translate TGSI_TEXTURE_x to VGAPU10_RESOURCE_DIMENSION_x.
*/
static unsigned
-pipe_texture_to_resource_dimension(unsigned target, bool msaa)
+tgsi_texture_to_resource_dimension(enum tgsi_texture_type target,
+ boolean is_array)
{
switch (target) {
- case PIPE_BUFFER:
+ case TGSI_TEXTURE_BUFFER:
return VGPU10_RESOURCE_DIMENSION_BUFFER;
- case PIPE_TEXTURE_1D:
+ case TGSI_TEXTURE_1D:
return VGPU10_RESOURCE_DIMENSION_TEXTURE1D;
- case PIPE_TEXTURE_2D:
- case PIPE_TEXTURE_RECT:
- return msaa ? VGPU10_RESOURCE_DIMENSION_TEXTURE2DMS
- : VGPU10_RESOURCE_DIMENSION_TEXTURE2D;
- case PIPE_TEXTURE_3D:
+ case TGSI_TEXTURE_2D:
+ case TGSI_TEXTURE_RECT:
+ return VGPU10_RESOURCE_DIMENSION_TEXTURE2D;
+ case TGSI_TEXTURE_3D:
return VGPU10_RESOURCE_DIMENSION_TEXTURE3D;
- case PIPE_TEXTURE_CUBE:
+ case TGSI_TEXTURE_CUBE:
+ return VGPU10_RESOURCE_DIMENSION_TEXTURECUBE;
+ case TGSI_TEXTURE_SHADOW1D:
+ return VGPU10_RESOURCE_DIMENSION_TEXTURE1D;
+ case TGSI_TEXTURE_SHADOW2D:
+ case TGSI_TEXTURE_SHADOWRECT:
+ return VGPU10_RESOURCE_DIMENSION_TEXTURE2D;
+ case TGSI_TEXTURE_1D_ARRAY:
+ case TGSI_TEXTURE_SHADOW1D_ARRAY:
+ return is_array ? VGPU10_RESOURCE_DIMENSION_TEXTURE1DARRAY
+ : VGPU10_RESOURCE_DIMENSION_TEXTURE1D;
+ case TGSI_TEXTURE_2D_ARRAY:
+ case TGSI_TEXTURE_SHADOW2D_ARRAY:
+ return is_array ? VGPU10_RESOURCE_DIMENSION_TEXTURE2DARRAY
+ : VGPU10_RESOURCE_DIMENSION_TEXTURE2D;
+ case TGSI_TEXTURE_SHADOWCUBE:
return VGPU10_RESOURCE_DIMENSION_TEXTURECUBE;
- case PIPE_TEXTURE_1D_ARRAY:
- return VGPU10_RESOURCE_DIMENSION_TEXTURE1DARRAY;
- case PIPE_TEXTURE_2D_ARRAY:
- return msaa ? VGPU10_RESOURCE_DIMENSION_TEXTURE2DMSARRAY
- : VGPU10_RESOURCE_DIMENSION_TEXTURE2DARRAY;
- case PIPE_TEXTURE_CUBE_ARRAY:
+ case TGSI_TEXTURE_2D_MSAA:
+ return VGPU10_RESOURCE_DIMENSION_TEXTURE2DMS;
+ case TGSI_TEXTURE_2D_ARRAY_MSAA:
+ return is_array ? VGPU10_RESOURCE_DIMENSION_TEXTURE2DMSARRAY
+ : VGPU10_RESOURCE_DIMENSION_TEXTURE2DMS;
+ case TGSI_TEXTURE_CUBE_ARRAY:
return VGPU10_RESOURCE_DIMENSION_TEXTURECUBEARRAY;
default:
assert(!"Unexpected resource type");
opcode0.value = 0;
opcode0.opcodeType = VGPU10_OPCODE_DCL_RESOURCE;
opcode0.resourceDimension =
- pipe_texture_to_resource_dimension(emit->key.tex[i].texture_target,
- emit->key.tex[i].texture_msaa);
+ tgsi_texture_to_resource_dimension(emit->sampler_target[i],
+ emit->key.tex[i].is_array);
operand0.value = 0;
operand0.numComponents = VGPU10_OPERAND_0_COMPONENT;
operand0.operandType = VGPU10_OPERAND_TYPE_RESOURCE;
STATIC_ASSERT(VGPU10_RETURN_TYPE_SINT == TGSI_RETURN_TYPE_SINT + 1);
STATIC_ASSERT(VGPU10_RETURN_TYPE_UINT == TGSI_RETURN_TYPE_UINT + 1);
STATIC_ASSERT(VGPU10_RETURN_TYPE_FLOAT == TGSI_RETURN_TYPE_FLOAT + 1);
- assert(emit->key.tex[i].return_type <= TGSI_RETURN_TYPE_FLOAT);
- rt = emit->key.tex[i].return_type + 1;
+ assert(emit->sampler_return_type[i] <= TGSI_RETURN_TYPE_FLOAT);
+ rt = emit->sampler_return_type[i] + 1;
#else
- switch (emit->key.tex[i].return_type) {
+ switch (emit->sampler_return_type[i]) {
case TGSI_RETURN_TYPE_UNORM: rt = VGPU10_RETURN_TYPE_UNORM; break;
case TGSI_RETURN_TYPE_SNORM: rt = VGPU10_RETURN_TYPE_SNORM; break;
case TGSI_RETURN_TYPE_SINT: rt = VGPU10_RETURN_TYPE_SINT; break;
static void
emit_instruction_op1(struct svga_shader_emitter_v10 *emit,
- unsigned opcode,
+ VGPU10_OPCODE_TYPE opcode,
const struct tgsi_full_dst_register *dst,
const struct tgsi_full_src_register *src,
boolean saturate)
static void
emit_instruction_op2(struct svga_shader_emitter_v10 *emit,
- unsigned opcode,
+ VGPU10_OPCODE_TYPE opcode,
const struct tgsi_full_dst_register *dst,
const struct tgsi_full_src_register *src1,
const struct tgsi_full_src_register *src2,
static void
emit_instruction_op3(struct svga_shader_emitter_v10 *emit,
- unsigned opcode,
+ VGPU10_OPCODE_TYPE opcode,
const struct tgsi_full_dst_register *dst,
const struct tgsi_full_src_register *src1,
const struct tgsi_full_src_register *src2,
unsigned i;
unsigned clip_plane_enable = emit->key.clip_plane_enable;
unsigned clip_dist_tmp_index = emit->clip_dist_tmp_index;
- unsigned num_written_clipdist = emit->info.num_written_clipdistance;
+ int num_written_clipdist = emit->info.num_written_clipdistance;
assert(emit->clip_dist_out_index != INVALID_INDEX);
assert(emit->clip_dist_tmp_index != INVALID_INDEX);
*/
emit->clip_dist_tmp_index = INVALID_INDEX;
- for (i = 0; i < 2 && num_written_clipdist; i++, num_written_clipdist-=4) {
+ for (i = 0; i < 2 && num_written_clipdist > 0; i++, num_written_clipdist-=4) {
tmp_clip_dist_src = make_src_temp_reg(clip_dist_tmp_index + i);
}
-/**
- * Emit code for TGSI_OPCODE_ABS instruction.
- */
-static boolean
-emit_abs(struct svga_shader_emitter_v10 *emit,
- const struct tgsi_full_instruction *inst)
-{
- /* dst = ABS(s0):
- * dst = abs(s0)
- * Translates into:
- * MOV dst, abs(s0)
- */
- struct tgsi_full_src_register abs_src0 = absolute_src(&inst->Src[0]);
-
- /* MOV dst, abs(s0) */
- emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &inst->Dst[0],
- &abs_src0, inst->Instruction.Saturate);
-
- return TRUE;
-}
-
-
/**
* Emit code for TGSI_OPCODE_ARL or TGSI_OPCODE_UARL instruction.
*/
{
unsigned index = inst->Dst[0].Register.Index;
struct tgsi_full_dst_register dst;
- unsigned opcode;
+ VGPU10_OPCODE_TYPE opcode;
assert(index < MAX_VGPU10_ADDR_REGS);
dst = make_dst_temp_reg(emit->address_reg_index[index]);
}
-/**
- * Emit code for TGSI_OPCODE_DP2A instruction.
- */
-static boolean
-emit_dp2a(struct svga_shader_emitter_v10 *emit,
- const struct tgsi_full_instruction *inst)
-{
- /* dst.x = src0.x * src1.x + src0.y * src1.y + src2.x
- * dst.y = src0.x * src1.x + src0.y * src1.y + src2.x
- * dst.z = src0.x * src1.x + src0.y * src1.y + src2.x
- * dst.w = src0.x * src1.x + src0.y * src1.y + src2.x
- * Translate into
- * MAD tmp.x, s0.y, s1.y, s2.x
- * MAD tmp.x, s0.x, s1.x, tmp.x
- * MOV dst.xyzw, tmp.xxxx
- */
- unsigned tmp = get_temp_index(emit);
- struct tgsi_full_src_register tmp_src = make_src_temp_reg(tmp);
- struct tgsi_full_dst_register tmp_dst = make_dst_temp_reg(tmp);
-
- struct tgsi_full_src_register tmp_src_xxxx =
- scalar_src(&tmp_src, TGSI_SWIZZLE_X);
- struct tgsi_full_dst_register tmp_dst_x =
- writemask_dst(&tmp_dst, TGSI_WRITEMASK_X);
-
- struct tgsi_full_src_register src0_xxxx =
- scalar_src(&inst->Src[0], TGSI_SWIZZLE_X);
- struct tgsi_full_src_register src0_yyyy =
- scalar_src(&inst->Src[0], TGSI_SWIZZLE_Y);
- struct tgsi_full_src_register src1_xxxx =
- scalar_src(&inst->Src[1], TGSI_SWIZZLE_X);
- struct tgsi_full_src_register src1_yyyy =
- scalar_src(&inst->Src[1], TGSI_SWIZZLE_Y);
- struct tgsi_full_src_register src2_xxxx =
- scalar_src(&inst->Src[2], TGSI_SWIZZLE_X);
-
- emit_instruction_op3(emit, VGPU10_OPCODE_MAD, &tmp_dst_x, &src0_yyyy,
- &src1_yyyy, &src2_xxxx, FALSE);
- emit_instruction_op3(emit, VGPU10_OPCODE_MAD, &tmp_dst_x, &src0_xxxx,
- &src1_xxxx, &tmp_src_xxxx, FALSE);
- emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &inst->Dst[0],
- &tmp_src_xxxx, inst->Instruction.Saturate);
-
- free_temp_indexes(emit);
-
- return TRUE;
-}
-
-
-/**
- * Emit code for TGSI_OPCODE_DPH instruction.
- */
-static boolean
-emit_dph(struct svga_shader_emitter_v10 *emit,
- const struct tgsi_full_instruction *inst)
-{
- /*
- * DP3 tmp, s0, s1
- * ADD dst, tmp, s1.wwww
- */
-
- struct tgsi_full_src_register s1_wwww =
- swizzle_src(&inst->Src[1], TGSI_SWIZZLE_W, TGSI_SWIZZLE_W,
- TGSI_SWIZZLE_W, TGSI_SWIZZLE_W);
-
- unsigned tmp = get_temp_index(emit);
- struct tgsi_full_src_register tmp_src = make_src_temp_reg(tmp);
- struct tgsi_full_dst_register tmp_dst = make_dst_temp_reg(tmp);
-
- /* DP3 tmp, s0, s1 */
- emit_instruction_op2(emit, VGPU10_OPCODE_DP3, &tmp_dst, &inst->Src[0],
- &inst->Src[1], FALSE);
-
- /* ADD dst, tmp, s1.wwww */
- emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &inst->Dst[0], &tmp_src,
- &s1_wwww, inst->Instruction.Saturate);
-
- free_temp_indexes(emit);
-
- return TRUE;
-}
-
-
/**
* Emit code for TGSI_OPCODE_DST instruction.
*/
}
-/**
- * Emit code for TGSI_OPCODE_SCS instruction.
- */
-static boolean
-emit_scs(struct svga_shader_emitter_v10 *emit,
- const struct tgsi_full_instruction *inst)
-{
- /* dst.x = cos(src.x)
- * dst.y = sin(src.x)
- * dst.z = 0.0
- * dst.w = 1.0
- */
- struct tgsi_full_dst_register dst_x =
- writemask_dst(&inst->Dst[0], TGSI_WRITEMASK_X);
- struct tgsi_full_dst_register dst_y =
- writemask_dst(&inst->Dst[0], TGSI_WRITEMASK_Y);
- struct tgsi_full_dst_register dst_zw =
- writemask_dst(&inst->Dst[0], TGSI_WRITEMASK_ZW);
-
- struct tgsi_full_src_register zero_one =
- make_immediate_reg_float4(emit, 0.0f, 0.0f, 0.0f, 1.0f);
-
- begin_emit_instruction(emit);
- emit_opcode(emit, VGPU10_OPCODE_SINCOS, inst->Instruction.Saturate);
- emit_dst_register(emit, &dst_y);
- emit_dst_register(emit, &dst_x);
- emit_src_register(emit, &inst->Src[0]);
- end_emit_instruction(emit);
-
- emit_instruction_op1(emit, VGPU10_OPCODE_MOV,
- &dst_zw, &zero_one, inst->Instruction.Saturate);
-
- return TRUE;
-}
-
-
/**
* Emit code for TGSI_OPCODE_SEQ (Set Equal) instruction.
*/
}
-/**
- * Emit code for TGSI_OPCODE_SUB instruction.
- */
-static boolean
-emit_sub(struct svga_shader_emitter_v10 *emit,
- const struct tgsi_full_instruction *inst)
-{
- /* dst = SUB(s0, s1):
- * dst = s0 - s1
- * Translates into:
- * ADD dst, s0, neg(s1)
- */
- struct tgsi_full_src_register neg_src1 = negate_src(&inst->Src[1]);
-
- /* ADD dst, s0, neg(s1) */
- emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &inst->Dst[0],
- &inst->Src[0], &neg_src1,
- inst->Instruction.Saturate);
-
- return TRUE;
-}
-
-
/**
* Emit a comparison instruction. The dest register will get
* 0 or ~0 values depending on the outcome of comparing src0 to src1.
struct tgsi_full_dst_register tmp_dst = make_dst_temp_reg(tmp);
struct tgsi_full_src_register scale_src = make_src_const_reg(scale_index);
- /* MUL tmp, coord, const[] */
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp_dst,
- coord, &scale_src, FALSE);
+ if (emit->key.tex[unit].texel_bias) {
+ /* to fix texture coordinate rounding issue, 0.0001 offset is
+ * been added. This fixes piglit test fbo-blit-scaled-linear. */
+ struct tgsi_full_src_register offset =
+ make_immediate_reg_float(emit, 0.0001f);
+
+ /* ADD tmp, coord, offset */
+ emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp_dst,
+ coord, &offset, FALSE);
+ /* MUL tmp, tmp, scale */
+ emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp_dst,
+ &tmp_src, &scale_src, FALSE);
+ }
+ else {
+ /* MUL tmp, coord, const[] */
+ emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp_dst,
+ coord, &scale_src, FALSE);
+ }
return tmp_src;
}
else {
*/
static void
emit_tex_compare_refcoord(struct svga_shader_emitter_v10 *emit,
- unsigned target,
+ enum tgsi_texture_type target,
const struct tgsi_full_src_register *coord)
{
struct tgsi_full_src_register coord_src_ref;
- unsigned component;
+ int component;
assert(tgsi_is_shadow_target(target));
- assert(target != TGSI_TEXTURE_SHADOWCUBE_ARRAY); /* XXX not implemented */
- if (target == TGSI_TEXTURE_SHADOW2D_ARRAY ||
- target == TGSI_TEXTURE_SHADOWCUBE)
- component = TGSI_SWIZZLE_W;
- else
- component = TGSI_SWIZZLE_Z;
+ component = tgsi_util_get_shadow_ref_src_index(target) % 4;
+ assert(component >= 0);
coord_src_ref = scalar_src(coord, component);
boolean swizzled;
boolean shadow_compare;
unsigned unit;
- unsigned texture_target; /**< TGSI_TEXTURE_x */
+ enum tgsi_texture_type texture_target; /**< TGSI_TEXTURE_x */
struct tgsi_full_src_register tmp_src;
struct tgsi_full_dst_register tmp_dst;
const struct tgsi_full_dst_register *inst_dst;
}
swz->inst_dst = &inst->Dst[0];
swz->coord_src = &inst->Src[0];
+
+ emit->fs.shadow_compare_units |= shadow_compare << unit;
}
assert(emit->unit == PIPE_SHADER_FRAGMENT);
- switch (swz->texture_target) {
- case TGSI_TEXTURE_SHADOW2D:
- case TGSI_TEXTURE_SHADOWRECT:
- case TGSI_TEXTURE_SHADOW1D_ARRAY:
- coord_src = scalar_src(swz->coord_src, TGSI_SWIZZLE_Z);
- break;
- case TGSI_TEXTURE_SHADOW1D:
- coord_src = scalar_src(swz->coord_src, TGSI_SWIZZLE_Y);
- break;
- case TGSI_TEXTURE_SHADOWCUBE:
- case TGSI_TEXTURE_SHADOW2D_ARRAY:
- coord_src = scalar_src(swz->coord_src, TGSI_SWIZZLE_W);
- break;
- default:
- assert(!"Unexpected texture target in end_tex_swizzle()");
- coord_src = scalar_src(swz->coord_src, TGSI_SWIZZLE_Z);
- }
+ int component =
+ tgsi_util_get_shadow_ref_src_index(swz->texture_target) % 4;
+ assert(component >= 0);
+ coord_src = scalar_src(swz->coord_src, component);
/* COMPARE tmp, coord, texel */
- /* XXX it would seem that the texel and coord arguments should
- * be transposed here, but piglit tests indicate otherwise.
- */
emit_comparison(emit, compare_func,
- &swz->tmp_dst, &texel_src, &coord_src);
+ &swz->tmp_dst, &coord_src, &texel_src);
/* AND dest, tmp, {1.0} */
begin_emit_instruction(emit);
unsigned swz_b = emit->key.tex[swz->unit].swizzle_b;
unsigned swz_a = emit->key.tex[swz->unit].swizzle_a;
unsigned writemask_0 = 0, writemask_1 = 0;
- boolean int_tex = is_integer_type(emit->key.tex[swz->unit].return_type);
+ boolean int_tex = is_integer_type(emit->sampler_return_type[swz->unit]);
/* Swizzle w/out zero/one terms */
struct tgsi_full_src_register src_swizzled =
swizzle_src(&swz->tmp_src,
- swz_r < PIPE_SWIZZLE_ZERO ? swz_r : PIPE_SWIZZLE_RED,
- swz_g < PIPE_SWIZZLE_ZERO ? swz_g : PIPE_SWIZZLE_GREEN,
- swz_b < PIPE_SWIZZLE_ZERO ? swz_b : PIPE_SWIZZLE_BLUE,
- swz_a < PIPE_SWIZZLE_ZERO ? swz_a : PIPE_SWIZZLE_ALPHA);
+ swz_r < PIPE_SWIZZLE_0 ? swz_r : PIPE_SWIZZLE_X,
+ swz_g < PIPE_SWIZZLE_0 ? swz_g : PIPE_SWIZZLE_Y,
+ swz_b < PIPE_SWIZZLE_0 ? swz_b : PIPE_SWIZZLE_Z,
+ swz_a < PIPE_SWIZZLE_0 ? swz_a : PIPE_SWIZZLE_W);
/* MOV dst, color(tmp).<swizzle> */
emit_instruction_op1(emit, VGPU10_OPCODE_MOV,
swz->inst_dst, &src_swizzled, FALSE);
/* handle swizzle zero terms */
- writemask_0 = (((swz_r == PIPE_SWIZZLE_ZERO) << 0) |
- ((swz_g == PIPE_SWIZZLE_ZERO) << 1) |
- ((swz_b == PIPE_SWIZZLE_ZERO) << 2) |
- ((swz_a == PIPE_SWIZZLE_ZERO) << 3));
+ writemask_0 = (((swz_r == PIPE_SWIZZLE_0) << 0) |
+ ((swz_g == PIPE_SWIZZLE_0) << 1) |
+ ((swz_b == PIPE_SWIZZLE_0) << 2) |
+ ((swz_a == PIPE_SWIZZLE_0) << 3));
+ writemask_0 &= swz->inst_dst->Register.WriteMask;
if (writemask_0) {
struct tgsi_full_src_register zero = int_tex ?
}
/* handle swizzle one terms */
- writemask_1 = (((swz_r == PIPE_SWIZZLE_ONE) << 0) |
- ((swz_g == PIPE_SWIZZLE_ONE) << 1) |
- ((swz_b == PIPE_SWIZZLE_ONE) << 2) |
- ((swz_a == PIPE_SWIZZLE_ONE) << 3));
+ writemask_1 = (((swz_r == PIPE_SWIZZLE_1) << 0) |
+ ((swz_g == PIPE_SWIZZLE_1) << 1) |
+ ((swz_b == PIPE_SWIZZLE_1) << 2) |
+ ((swz_a == PIPE_SWIZZLE_1) << 3));
+ writemask_1 &= swz->inst_dst->Register.WriteMask;
if (writemask_1) {
struct tgsi_full_src_register one = int_tex ?
/* SAMPLE dst, coord(s0), resource, sampler */
begin_emit_instruction(emit);
+ /* NOTE: for non-fragment shaders, we should use VGPU10_OPCODE_SAMPLE_L
+ * with LOD=0. But our virtual GPU accepts this as-is.
+ */
emit_sample_opcode(emit, VGPU10_OPCODE_SAMPLE,
inst->Instruction.Saturate, offsets);
emit_dst_register(emit, get_tex_swizzle_dst(&swz_info));
const struct tgsi_full_instruction *inst)
{
const unsigned unit = inst->Src[1].Register.Index;
- const unsigned target = inst->Texture.Texture;
+ const enum tgsi_texture_type target = inst->Texture.Texture;
boolean valid = TRUE;
if (tgsi_is_shadow_target(target) &&
- is_integer_type(emit->key.tex[unit].return_type)) {
+ is_integer_type(emit->sampler_return_type[unit])) {
debug_printf("Invalid SAMPLE_C with an integer texture!\n");
valid = FALSE;
}
const struct tgsi_full_instruction *inst)
{
const uint unit = inst->Src[1].Register.Index;
- unsigned target = inst->Texture.Texture;
- unsigned opcode;
+ const enum tgsi_texture_type target = inst->Texture.Texture;
+ VGPU10_OPCODE_TYPE opcode;
struct tgsi_full_src_register coord;
int offsets[3];
struct tex_swizzle_info swz_info;
const struct tgsi_full_instruction *inst)
{
const uint unit = inst->Src[1].Register.Index;
- unsigned target = inst->Texture.Texture;
- unsigned opcode;
+ const enum tgsi_texture_type target = inst->Texture.Texture;
+ VGPU10_OPCODE_TYPE opcode;
int offsets[3];
unsigned tmp = get_temp_index(emit);
struct tgsi_full_src_register tmp_src = make_src_temp_reg(tmp);
begin_emit_instruction(emit);
if (tgsi_is_shadow_target(target))
+ /* NOTE: for non-fragment shaders, we should use
+ * VGPU10_OPCODE_SAMPLE_C_LZ, but our virtual GPU accepts this as-is.
+ */
opcode = VGPU10_OPCODE_SAMPLE_C;
else
opcode = VGPU10_OPCODE_SAMPLE;
}
-/*
- * Emit code for TGSI_OPCODE_XPD instruction.
- */
-static boolean
-emit_xpd(struct svga_shader_emitter_v10 *emit,
- const struct tgsi_full_instruction *inst)
-{
- /* dst.x = src0.y * src1.z - src1.y * src0.z
- * dst.y = src0.z * src1.x - src1.z * src0.x
- * dst.z = src0.x * src1.y - src1.x * src0.y
- * dst.w = 1
- */
- struct tgsi_full_src_register s0_xxxx =
- scalar_src(&inst->Src[0], TGSI_SWIZZLE_X);
- struct tgsi_full_src_register s0_yyyy =
- scalar_src(&inst->Src[0], TGSI_SWIZZLE_Y);
- struct tgsi_full_src_register s0_zzzz =
- scalar_src(&inst->Src[0], TGSI_SWIZZLE_Z);
-
- struct tgsi_full_src_register s1_xxxx =
- scalar_src(&inst->Src[1], TGSI_SWIZZLE_X);
- struct tgsi_full_src_register s1_yyyy =
- scalar_src(&inst->Src[1], TGSI_SWIZZLE_Y);
- struct tgsi_full_src_register s1_zzzz =
- scalar_src(&inst->Src[1], TGSI_SWIZZLE_Z);
-
- unsigned tmp1 = get_temp_index(emit);
- struct tgsi_full_src_register tmp1_src = make_src_temp_reg(tmp1);
- struct tgsi_full_dst_register tmp1_dst = make_dst_temp_reg(tmp1);
-
- unsigned tmp2 = get_temp_index(emit);
- struct tgsi_full_src_register tmp2_src = make_src_temp_reg(tmp2);
- struct tgsi_full_dst_register tmp2_dst = make_dst_temp_reg(tmp2);
- struct tgsi_full_src_register neg_tmp2_src = negate_src(&tmp2_src);
-
- unsigned tmp3 = get_temp_index(emit);
- struct tgsi_full_src_register tmp3_src = make_src_temp_reg(tmp3);
- struct tgsi_full_dst_register tmp3_dst = make_dst_temp_reg(tmp3);
- struct tgsi_full_dst_register tmp3_dst_x =
- writemask_dst(&tmp3_dst, TGSI_WRITEMASK_X);
- struct tgsi_full_dst_register tmp3_dst_y =
- writemask_dst(&tmp3_dst, TGSI_WRITEMASK_Y);
- struct tgsi_full_dst_register tmp3_dst_z =
- writemask_dst(&tmp3_dst, TGSI_WRITEMASK_Z);
- struct tgsi_full_dst_register tmp3_dst_w =
- writemask_dst(&tmp3_dst, TGSI_WRITEMASK_W);
-
- /* Note: we put all the intermediate computations into tmp3 in case
- * the XPD dest register is that same as one of the src regs (in which
- * case we could clobber a src reg before we're done with it) .
- *
- * Note: we could get by with just one temp register instead of three
- * since we're doing scalar operations and there's enough room in one
- * temp for everything.
- */
-
- /* MUL tmp1, src0.y, src1.z */
- /* MUL tmp2, src1.y, src0.z */
- /* ADD tmp3.x, tmp1, -tmp2 */
- if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp1_dst,
- &s0_yyyy, &s1_zzzz, FALSE);
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp2_dst,
- &s1_yyyy, &s0_zzzz, FALSE);
- emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp3_dst_x,
- &tmp1_src, &neg_tmp2_src, FALSE);
- }
-
- /* MUL tmp1, src0.z, src1.x */
- /* MUL tmp2, src1.z, src0.x */
- /* ADD tmp3.y, tmp1, -tmp2 */
- if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp1_dst, &s0_zzzz,
- &s1_xxxx, FALSE);
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp2_dst, &s1_zzzz,
- &s0_xxxx, FALSE);
- emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp3_dst_y,
- &tmp1_src, &neg_tmp2_src, FALSE);
- }
-
- /* MUL tmp1, src0.x, src1.y */
- /* MUL tmp2, src1.x, src0.y */
- /* ADD tmp3.z, tmp1, -tmp2 */
- if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Z) {
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp1_dst, &s0_xxxx,
- &s1_yyyy, FALSE);
- emit_instruction_op2(emit, VGPU10_OPCODE_MUL, &tmp2_dst, &s1_xxxx,
- &s0_yyyy, FALSE);
- emit_instruction_op2(emit, VGPU10_OPCODE_ADD, &tmp3_dst_z,
- &tmp1_src, &neg_tmp2_src, FALSE);
- }
-
- /* MOV tmp3.w, 1.0 */
- if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_W) {
- struct tgsi_full_src_register one =
- make_immediate_reg_float(emit, 1.0f);
-
- emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &tmp3_dst_w, &one, FALSE);
- }
-
- /* MOV dst, tmp3 */
- emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &inst->Dst[0], &tmp3_src,
- inst->Instruction.Saturate);
-
-
- free_temp_indexes(emit);
-
- return TRUE;
-}
-
-
/**
* Emit code for TGSI_OPCODE_TXD (explicit derivatives)
*/
const struct tgsi_full_instruction *inst)
{
const uint unit = inst->Src[3].Register.Index;
- unsigned target = inst->Texture.Texture;
+ const enum tgsi_texture_type target = inst->Texture.Texture;
int offsets[3];
struct tgsi_full_src_register coord;
struct tex_swizzle_info swz_info;
const struct tgsi_full_instruction *inst)
{
const uint unit = inst->Src[1].Register.Index;
- const unsigned msaa = emit->key.tex[unit].texture_msaa;
+ const boolean msaa = tgsi_is_msaa_target(inst->Texture.Texture);
int offsets[3];
struct tex_swizzle_info swz_info;
emit_txl_txb(struct svga_shader_emitter_v10 *emit,
const struct tgsi_full_instruction *inst)
{
- unsigned target = inst->Texture.Texture;
- unsigned opcode, unit;
+ const enum tgsi_texture_type target = inst->Texture.Texture;
+ VGPU10_OPCODE_TYPE opcode;
+ unsigned unit;
int offsets[3];
struct tgsi_full_src_register coord, lod_bias;
struct tex_swizzle_info swz_info;
{
const uint unit = inst->Src[1].Register.Index;
- if (emit->key.tex[unit].texture_target == PIPE_BUFFER) {
+ if (emit->sampler_target[unit] == TGSI_TEXTURE_BUFFER) {
/* RESINFO does not support querying texture buffers, so we instead
* store texture buffer sizes in shader constants, then copy them to
* implement TXQ instead of emitting RESINFO.
emit_simple(struct svga_shader_emitter_v10 *emit,
const struct tgsi_full_instruction *inst)
{
- const unsigned opcode = inst->Instruction.Opcode;
+ const enum tgsi_opcode opcode = inst->Instruction.Opcode;
const struct tgsi_opcode_info *op = tgsi_get_opcode_info(opcode);
unsigned i;
begin_emit_instruction(emit);
- emit_opcode(emit, translate_opcode(inst->Instruction.Opcode),
- inst->Instruction.Saturate);
+ emit_opcode(emit, translate_opcode(opcode), inst->Instruction.Saturate);
for (i = 0; i < op->num_dst; i++) {
emit_dst_register(emit, &inst->Dst[i]);
}
}
+/**
+ * We only special case the MOV instruction to try to detect constant
+ * color writes in the fragment shader.
+ */
+static boolean
+emit_mov(struct svga_shader_emitter_v10 *emit,
+ const struct tgsi_full_instruction *inst)
+{
+ const struct tgsi_full_src_register *src = &inst->Src[0];
+ const struct tgsi_full_dst_register *dst = &inst->Dst[0];
+
+ if (emit->unit == PIPE_SHADER_FRAGMENT &&
+ dst->Register.File == TGSI_FILE_OUTPUT &&
+ dst->Register.Index == 0 &&
+ src->Register.File == TGSI_FILE_CONSTANT &&
+ !src->Register.Indirect) {
+ emit->constant_color_output = TRUE;
+ }
+
+ return emit_simple(emit, inst);
+}
+
+
/**
* Emit a simple VGPU10 instruction which writes to multiple dest registers,
* where TGSI only uses one dest register.
unsigned dst_count,
unsigned dst_index)
{
- const unsigned opcode = inst->Instruction.Opcode;
+ const enum tgsi_opcode opcode = inst->Instruction.Opcode;
const struct tgsi_opcode_info *op = tgsi_get_opcode_info(opcode);
unsigned i;
begin_emit_instruction(emit);
- emit_opcode(emit, translate_opcode(inst->Instruction.Opcode),
- inst->Instruction.Saturate);
+ emit_opcode(emit, translate_opcode(opcode), inst->Instruction.Saturate);
for (i = 0; i < dst_count; i++) {
if (i == dst_index) {
unsigned inst_number,
const struct tgsi_full_instruction *inst)
{
- const unsigned opcode = inst->Instruction.Opcode;
+ const enum tgsi_opcode opcode = inst->Instruction.Opcode;
switch (opcode) {
case TGSI_OPCODE_ADD:
case TGSI_OPCODE_MAD:
case TGSI_OPCODE_MAX:
case TGSI_OPCODE_MIN:
- case TGSI_OPCODE_MOV:
case TGSI_OPCODE_MUL:
case TGSI_OPCODE_NOP:
case TGSI_OPCODE_NOT:
/* simple instructions */
return emit_simple(emit, inst);
-
+ case TGSI_OPCODE_MOV:
+ return emit_mov(emit, inst);
case TGSI_OPCODE_EMIT:
return emit_vertex(emit, inst);
case TGSI_OPCODE_ENDPRIM:
return emit_endprim(emit, inst);
- case TGSI_OPCODE_ABS:
- return emit_abs(emit, inst);
case TGSI_OPCODE_IABS:
return emit_iabs(emit, inst);
case TGSI_OPCODE_ARL:
return emit_cmp(emit, inst);
case TGSI_OPCODE_COS:
return emit_sincos(emit, inst);
- case TGSI_OPCODE_DP2A:
- return emit_dp2a(emit, inst);
- case TGSI_OPCODE_DPH:
- return emit_dph(emit, inst);
case TGSI_OPCODE_DST:
return emit_dst(emit, inst);
case TGSI_OPCODE_EX2:
return emit_rsq(emit, inst);
case TGSI_OPCODE_SAMPLE:
return emit_sample(emit, inst);
- case TGSI_OPCODE_SCS:
- return emit_scs(emit, inst);
case TGSI_OPCODE_SEQ:
return emit_seq(emit, inst);
case TGSI_OPCODE_SGE:
return emit_ssg(emit, inst);
case TGSI_OPCODE_ISSG:
return emit_issg(emit, inst);
- case TGSI_OPCODE_SUB:
- return emit_sub(emit, inst);
case TGSI_OPCODE_TEX:
return emit_tex(emit, inst);
case TGSI_OPCODE_TXP:
return emit_txq(emit, inst);
case TGSI_OPCODE_UIF:
return emit_if(emit, inst);
- case TGSI_OPCODE_XPD:
- return emit_xpd(emit, inst);
case TGSI_OPCODE_UMUL_HI:
case TGSI_OPCODE_IMUL_HI:
case TGSI_OPCODE_UDIV:
while (adjust_mask) {
unsigned index = u_bit_scan(&adjust_mask);
+
+ /* skip the instruction if this vertex attribute is not being used */
+ if (emit->info.input_usage_mask[index] == 0)
+ continue;
+
unsigned tmp = emit->vs.adjusted_input[index];
struct tgsi_full_src_register input_src =
make_src_reg(TGSI_FILE_INPUT, index);
emit->common_immediate_pos[n++] =
alloc_immediate_float4(emit, 0.0f, 1.0f, 0.5f, -1.0f);
- emit->common_immediate_pos[n++] =
- alloc_immediate_float4(emit, 128.0f, -128.0f, 2.0f, 3.0f);
+ if (emit->info.opcode_count[TGSI_OPCODE_LIT] > 0) {
+ emit->common_immediate_pos[n++] =
+ alloc_immediate_float4(emit, 128.0f, -128.0f, 0.0f, 0.0f);
+ }
emit->common_immediate_pos[n++] =
alloc_immediate_int4(emit, 0, 1, 0, -1);
if (emit->key.vs.attrib_puint_to_snorm) {
emit->common_immediate_pos[n++] =
- alloc_immediate_float4(emit, -2.0f, -2.0f, -2.0f, -1.66666f);
+ alloc_immediate_float4(emit, -2.0f, 2.0f, 3.0f, -1.66666f);
}
if (emit->key.vs.attrib_puint_to_uscaled) {
alloc_immediate_int4(emit, 22, 30, 0, 0);
}
- assert(n <= Elements(emit->common_immediate_pos));
+ unsigned i;
+
+ for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+ if (emit->key.tex[i].texel_bias) {
+ /* Replace 0.0f if more immediate float value is needed */
+ emit->common_immediate_pos[n++] =
+ alloc_immediate_float4(emit, 0.0001f, 0.0f, 0.0f, 0.0f);
+ break;
+ }
+ }
+
+ assert(n <= ARRAY_SIZE(emit->common_immediate_pos));
emit->num_common_immediates = n;
}
}
+/**
+ * The device has no direct support for the pipe_blend_state::alpha_to_one
+ * option so we implement it here with shader code.
+ *
+ * Note that this is kind of pointless, actually. Here we're clobbering
+ * the alpha value with 1.0. So if alpha-to-coverage is enabled, we'll wind
+ * up with 100% coverage. That's almost certainly not what the user wants.
+ * The work-around is to add extra shader code to compute coverage from alpha
+ * and write it to the coverage output register (if the user's shader doesn't
+ * do so already). We'll probably do that in the future.
+ */
+static void
+emit_alpha_to_one_instructions(struct svga_shader_emitter_v10 *emit,
+ unsigned fs_color_tmp_index)
+{
+ struct tgsi_full_src_register one = make_immediate_reg_float(emit, 1.0f);
+ unsigned i;
+
+ /* Note: it's not 100% clear from the spec if we're supposed to clobber
+ * the alpha for all render targets. But that's what NVIDIA does and
+ * that's what Piglit tests.
+ */
+ for (i = 0; i < emit->fs.num_color_outputs; i++) {
+ struct tgsi_full_dst_register color_dst;
+
+ if (fs_color_tmp_index != INVALID_INDEX && i == 0) {
+ /* write to the temp color register */
+ color_dst = make_dst_temp_reg(fs_color_tmp_index);
+ }
+ else {
+ /* write directly to the color[i] output */
+ color_dst = make_dst_output_reg(emit->fs.color_out_index[i]);
+ }
+
+ color_dst = writemask_dst(&color_dst, TGSI_WRITEMASK_W);
+
+ emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &color_dst, &one, FALSE);
+ }
+}
+
+
/**
* Emit alpha test code. This compares TEMP[fs_color_tmp_index].w
* against the alpha reference value and discards the fragment if the
emit_src_register(emit, &tmp_src_x);
end_emit_instruction(emit);
- /* If we don't need to broadcast the color below, emit final color here */
+ /* If we don't need to broadcast the color below, emit the final color here.
+ */
if (emit->key.fs.write_color0_to_n_cbufs <= 1) {
/* MOV output.color, tempcolor */
emit_instruction_op1(emit, VGPU10_OPCODE_MOV, &color_dst,
/**
* Emit instructions for writing a single color output to multiple
* color buffers.
- * This is used when the TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS
+ * This is used when the TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS (or
+ * when key.fs.white_fragments is true).
* property is set and the number of render targets is greater than one.
* \param fs_color_tmp_index index of the temp register that holds the
* color to broadcast.
{
const unsigned n = emit->key.fs.write_color0_to_n_cbufs;
unsigned i;
- struct tgsi_full_src_register color_src =
- make_src_temp_reg(fs_color_tmp_index);
+ struct tgsi_full_src_register color_src;
+
+ if (emit->key.fs.white_fragments) {
+ /* set all color outputs to white */
+ color_src = make_immediate_reg_float(emit, 1.0f);
+ }
+ else {
+ /* set all color outputs to TEMP[fs_color_tmp_index] */
+ assert(fs_color_tmp_index != INVALID_INDEX);
+ color_src = make_src_temp_reg(fs_color_tmp_index);
+ }
assert(emit->unit == PIPE_SHADER_FRAGMENT);
- assert(n > 1);
for (i = 0; i < n; i++) {
unsigned output_reg = emit->fs.color_out_index[i];
else if (emit->unit == PIPE_SHADER_FRAGMENT) {
const unsigned fs_color_tmp_index = emit->fs.color_tmp_index;
+ assert(!(emit->key.fs.white_fragments &&
+ emit->key.fs.write_color0_to_n_cbufs == 0));
+
/* We no longer want emit_dst_register() to substitute the
* temporary fragment color register for the real color output.
*/
emit->fs.color_tmp_index = INVALID_INDEX;
+ if (emit->key.fs.alpha_to_one) {
+ emit_alpha_to_one_instructions(emit, fs_color_tmp_index);
+ }
if (emit->key.fs.alpha_func != SVGA3D_CMP_ALWAYS) {
emit_alpha_test_instructions(emit, fs_color_tmp_index);
}
- if (emit->key.fs.write_color0_to_n_cbufs > 1) {
+ if (emit->key.fs.write_color0_to_n_cbufs > 1 ||
+ emit->key.fs.white_fragments) {
emit_broadcast_color_instructions(emit, fs_color_tmp_index);
}
}
tgsi_dump(tokens,0);
}
- new_tokens = util_pstipple_create_fragment_shader(tokens, &unit, 0);
+ new_tokens = util_pstipple_create_fragment_shader(tokens, &unit, 0,
+ TGSI_FILE_INPUT);
emit->fs.pstipple_sampler_unit = unit;
/* Setup texture state for stipple */
- emit->key.tex[unit].texture_target = PIPE_TEXTURE_2D;
+ emit->sampler_target[unit] = TGSI_TEXTURE_2D;
emit->key.tex[unit].swizzle_r = TGSI_SWIZZLE_X;
emit->key.tex[unit].swizzle_g = TGSI_SWIZZLE_Y;
emit->key.tex[unit].swizzle_b = TGSI_SWIZZLE_Z;
svga_tgsi_vgpu10_translate(struct svga_context *svga,
const struct svga_shader *shader,
const struct svga_compile_key *key,
- unsigned unit)
+ enum pipe_shader_type unit)
{
struct svga_shader_variant *variant = NULL;
struct svga_shader_emitter_v10 *emit;
/* These two flags cannot be used together */
assert(key->vs.need_prescale + key->vs.undo_viewport <= 1);
+ SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_TGSIVGPU10TRANSLATE);
/*
* Setup the code emitter
*/
emit = alloc_emitter();
if (!emit)
- return NULL;
+ goto done;
emit->unit = unit;
emit->key = *key;
/*
* Create, initialize the 'variant' object.
*/
- variant = CALLOC_STRUCT(svga_shader_variant);
+ variant = svga_new_shader_variant(svga);
if (!variant)
goto cleanup;
variant->pstipple_sampler_unit = emit->fs.pstipple_sampler_unit;
+ /* If there was exactly one write to a fragment shader output register
+ * and it came from a constant buffer, we know all fragments will have
+ * the same color (except for blending).
+ */
+ variant->constant_color_output =
+ emit->constant_color_output && emit->num_output_writes == 1;
+
/** keep track in the variant if flat interpolation is used
* for any of the varyings.
*/
variant->uses_flat_interp = emit->uses_flat_interp;
+ variant->fs_shadow_compare_units = emit->fs.shadow_compare_units;
+
if (tokens != shader->tokens) {
tgsi_free_tokens(tokens);
}
cleanup:
free_emitter(emit);
+done:
+ SVGA_STATS_TIME_POP(svga_sws(svga));
return variant;
}